Your $c_n(\lambda)$ is closely related to what [Billey, Konvalinka, Swanson 2020](https://arxiv.org/abs/1905.00975) call the aft of a partition:
$$ \text{aft}(\lambda) = | \lambda | - \max\{\lambda_1, \lambda'_1\}$$
where $\lambda'$ is the conjugate of $\lambda$, so $\lambda'_1 = \ell(\lambda)$.

Following up on Martin Rubey's comment, your Question 1 is true: For $c_n(\lambda) > n/2$, we have $\lambda \ne \lambda'$ since $\lambda_1 = \lambda'_1$ would make $2\lambda_1 > n$, a contradiction.  Therefore, $\lambda$ is not self-conjugate and each partition $\mu \vdash n - \lambda_1$ is counted twice in the coefficient of $t^{\lambda_1}$, once for the partition of $n$ consisting of $\lambda_1$ followed by $\mu$ and once for the conjugate of that partition of $n$.

For your Question 2, a generating function in terms of Gaussian binomial coefficients is given in the related OEIS entry [A338621](https://oeis.org/A338621).